ByteDance plans to invest 160 billion RMB in AI infrastructure by 2026, with “half of the budget” earmarked for purchasing NVIDIA H200 chips, striving to maintain AI competitiveness under the blockades of the Trump 2.0 era. (Previous summary: The US version of TikTok is officially sold: three major US capital groups including Oracle take over, ByteDance retains the brain, did Trump win?) (Background Supplement: “ByteDance's Guo Yu, who retired at 28, speaks the truth at 34: Financial freedom is not true freedom”)

Table of Contents

• The capital scale is only 1/10 of Silicon Valley.
• 85 billion RMB buying H200
• Doubao and overseas data center dual-line computing power

According to the Financial Times, in the winter of 2025, one year after Trump returns to the White House and as Washington's blockade against Chinese tech companies intensifies, ByteDance has issued a check for 160 billion RMB for AI infrastructure next year, attempting to preserve its survival space amid a possible “chip blockade” in the future. This budget, equivalent to 23 billion USD, is precisely the bet that TikTok's parent company is placing on AI.

The capital scale is only 1/10 of Silicon Valley.

According to the latest leaked internal documents, ByteDance will increase its capital expenditure from 150 billion yuan to 160 billion yuan in 2026, reflecting an annual increase of about 7%. Compared to the more than 300 billion dollars in Capex combined for the four major cloud giants in the United States in 2025, ByteDance's amount is less than one-tenth of its competitors. However, in the context of the Chinese market, this is already the largest capacity available to a single private enterprise, highlighting its anxiety over the AI cost curve.

Funding is not the only threshold; obtaining chips is the core issue. Without high-performance GPUs, money cannot be converted into computing power; without computing power, model iterations will come to a halt. ByteDance views this 160 billion RMB as a “survival policy,” aiming to shorten the R&D cycle gap with Microsoft and Meta.

850 billion RMB buying spree H200

About half of it, reaching 85 billion yuan (approximately 12 billion USD), is specifically allocated for purchasing advanced AI processors. According to external analysis, ByteDance has placed an order with Nvidia for 20,000 H200 chips, with the unit price still under negotiation.

However, due to the latest export restrictions imposed by the Trump administration, the batch of orders is currently in a “pending approval” status and may need to be delivered through methods such as specification downgrade or delivery to a third location. Earlier, it was rumored that the procurement scale of “850 billion RMB” had been clarified as a misunderstanding, but the figure of 85 billion itself remains the largest GPU procurement case ever by a Chinese company.

Doubao and overseas data center dual line pulling computing power

Where the money is spent depends on the application push. ByteDance's internal multimodal large model has supported the domestic chatbot “Doubao,” which has seen exponential user growth in recent years, with inference usage nearly matching that of short video frameworks. To reduce the risks associated with limited domestic cloud services, ByteDance is simultaneously setting up a new generation of data centers in Southeast Asia and Europe for model training and TikTok recommendation algorithm iteration. Hardware is first purchased with investment, so that applications can potentially iterate continuously.

Against the backdrop of rapidly escalating semiconductor export controls, ByteDance's significant investments appear more like a defensive stockpiling strategy, possibly betting on a brief policy window. According to the Financial Times, ByteDance's executives have assessed that “there will be fewer chips available for purchase next year, and it will be harder to spend money.” This indicates that time has become the factor with the highest implicit cost.